Professional Documents
Culture Documents
http://www.continuummechanics.org/polardec...
Polar
Deformation Gradients
Decomposition
Introduction
The polar decomposition concept was
introduced on the previous
deformation gradient page. In it, we saw
through example that F can be written
as either R U or V R. In each case,
R is the rotation matrix, and U and V
are symmetric matrices describing the
deformations.
The purpose of this page is to show how
to compute a polar decomposition in the
general 3-D case. That process will
introduce another new concept: the
square root of a matrix. But rst...
FT F
But as with
1 of 10
Q, here RT
Ref:
http://commons.wikimedia.org
/wiki/File:Polar_decomposition_of_F.png
(R U)T (R U)
is the inverse of
UT RT R U
Polar Decomposition
http://www.continuummechanics.org/polardec...
RT R
And
R1 R
FT F reduces to
FT F
(R U)T (R U)
UT RT R U
UT U
The result is called the "Right Cauchy-Green Deformation Tensor," and sometimes
represented by C, but I don't like this because it hides the true physics behind the letter
T
which is U U. So I won't be using it.
F FT
F FT
using
C,
F = V R to obtain
(V R) (V R)T
V R RT VT
V VT
And this result is called the "Left Cauchy-Green Deformation Tensor," and sometimes
represented by B , but I don't like this because it also hides the true physics behind the letter
B , which is V VT . So I won't be using it either.
The surprising result here is that the rotation matrix, R, has been eliminated from the problem in each
T
T
case. But a new challenge has been introduced, that of nding U from (U U) and V from (V V ).
This is the source of the need to compute square roots of matrices. Recall that U and V are both
T
T
T
T
symmetric, so U = U and U U = U U . Likewise V = V and V V = V V . Finally, U U is
2
2
sometimes written as U and V V is sometimes written as V .
Determination of R & U
Recall the example from the previous page where the object is
transformed from a square to the position shown in the gure.
The equations to do this are
x = 1.300X 0.375 Y
y = 0.750X + 0.650 Y
and the deformation gradient is
F=
So
0.375
0.650 ]
FT F is
FT F
2 of 10
1.300
[ 0.750
1.300
[ 0.375
0.750
1.300
0.650 ] [ 0.750
0.375
0.650 ]
2.250
[ 0.000
0.000
0.563 ]
UT U
Polar Decomposition
http://www.continuummechanics.org/polardec...
This is the simplest of all scenarios - a diagonal matrix. In this case, the
square-roots of the diagonal values.
2.25
[ 0.0
0.0
]
0.563
1.5
[ 0.0
which is exactly the U matrix result from the previous page. Once
multiply F through by the inverse of U to get R.
F U1
And the inverse of
R U U1
0.0
0.75 ]
So
0.667
[ 0.0
0.0
1.333 ]
R is
R
F U1
1.300
[ 0.750
0.375
0.667
0.650 ] [ 0.0
0.0
1.333 ]
0.866
[ 0.500
0.500
0.866 ]
This concludes this example of determining R and U from a deformation matrix, F. However, it was
deceptively simple in this case because U was a diagonal matrix, i.e., no shear. The next example will
show the more typical, and much more complex case of performing a polar decomposition when shear
is present in the problem.
Determination of V & R
x = 1.300X 0.375 Y
y = 0.750X + 0.650 Y
This time, performing the
F FT
1.300
[ 0.750
0.750
0.650 ]
1.831
[ 0.731
0.731
0.985 ]
V VT
This is the more common situation in which the resulting matrix is not diagonal. This greatly
complicates the process of nding its square root. As was the case with U, recall that V is also
T
T
symmetric, so V = V and V V = V V .
3 of 10
Polar Decomposition
http://www.continuummechanics.org/polardec...
The following process of nding square roots of matrices only applies to symmetric ones. Otherwise,
the square root will contain imaginary numbers that are of little use in the non-imaginary world of
continuum mechanics.
The steps are as follows:
1. Transform the symmetric matrix to its principal orientation
2. Take the square roots of the diagonal components
3. Rotate back to the original orientation
Step 1 is relatively easy to accomplish for the 2-D example here. (A 3-D example will be presented
later.) The rotation angle is
2a12
2 0.731
tan 2 = (
=(
)
a11 a22
1.831 0.985 )
Applying the 30 rotation to
cos 30
[ sin 30
V VT
is done as follows
sin 30
1.831
][
cos 30
0.731
0.731
cos 30
0.985 ] [ sin 30
sin 30
2.250
=
]
[ 0.0
cos 30
2.25
[ 0.0
= 30
0.0
]
0.563
V VT
1.5
[ 0.0
0.0
0.563 ]
0.0
0.75 ]
The nal step is to transform this result back by -30 to the original orientation.
cos(-30 )
[ sin(-30 )
sin(-30 )
1.5
][
cos(-30 )
0.0
0.0
cos(-30 )
0.75 ] [ sin(-30 )
sin(-30 )
1.313
=
]
[ 0.325
cos(-30 )
0.325
0.938 ]
VV
1.313
[ 0.325
0.325
1.313
0.938 ] [ 0.325
0.325
1.831
=
0.938 ] [ 0.731
0.731
0.985 ]
So this is correct.
Note that one thing that is not correct is to square each individual term of a matrix.
[ 0.325
1.3132
4 of 10
0.938 ]
0.3252
2
1.724
[ 0.106
0.106
1.831
0.880 ] [ 0.731
0.731
0.985 ]
Polar Decomposition
Once
http://www.continuummechanics.org/polardec...
The inverse of
R is
And
V1 V R
V1 F
0.833
[ 0.289
0.833
[ 0.289
0.289
1.166 ]
0.289
1.300
1.166 ] [ 0.750
0.375
0.650 ]
0.866
[ 0.500
0.500
0.866 ]
Review
So we have seen that given a simple coordinate mapping of
x = 1.300X 0.375 Y
y = 0.750X + 0.650 Y
The deformation gradient is
F=
1.300
[ 0.750
0.375
0.650 ]
RU
0.866
[ 0.500
0.500
1.50
]
[
0.866
0.0
0.0
0.75 ]
Or alternatively
VR
1.313
[ 0.325
0.325
0.866
0.938 ] [ 0.500
0.500
0.866 ]
In each case, the rotation is the same (because it must be). It's important to understand that the
deformations are the same also. They just appear to be dierent because one is imposed before the
object is rotated, the other after. To me, the advantage of the R U decomposition, is that one can
easily identify the deformations because they are applied to the underformed, unrotated,
un-anythinged object. So one does not need to know how much the object rotated before knowing
5 of 10
Polar Decomposition
http://www.continuummechanics.org/polardec...
3-D Example
Here is a full-blown 3-D example. An object deforms
according to
x = X + 2Y sin t + 0.5Z
y = 0.333X + Y Z sin t
z = X 2 sin 2t + 1.5Z
so the deformation gradient is
1
xi
F=
= 0.333
Xj
2X sin 2t
And the question is, "What are the conditions at
2 sin t
1
0
0.5
sin t
1.5
1
xi
F=
= 0.333
Xj
0.959
Begin the process by premultiplying
to multiply the matrices.
0.495
1
0
0.5
0.247
1.5
2.031
FT F = 0.162
2.021
0.162
1.245
0
2.021
0 = UT U
2.561
The next step is to nd an orientation (i.e., a transformed coordinate system) that makes the result a
diagonal matrix so that the square roots can be taken. This page on eigenvalues and eigenvectors can
do that.
6 of 10
Polar Decomposition
http://www.continuummechanics.org/polardec...
0.243
U2 = 0
0
and the transformation matrix,
UT U are
0
1.256
0
0
4.338
Q = 0.068
0.660
0.121
0.992
0.035
0.652
0.106
0.751
The next step is to take the square root of the diagonal elements of
0.493
U = 0
0
0
1.121
0
U2
to get
0
2.083
and transform it back to the original coordinate system. There is in fact a relatively easy way to do
this. It is
U = QT U Q
Multiplying this out gives
7 of 10
Polar Decomposition
http://www.continuummechanics.org/polardec...
1.188
U = 0.079
0.783
This shows that there is a lot of stretching in the
a lot of shear in the X Z direction.
The inverse of
1.349
= 0.112
0.759
and
directions, and
0.112
0.908
0.079
0.759
0.079
1.144
R = F U1
R = F U1
0.914
= 0.374
0.156
8 of 10
0.783
0.024
1.396
and nally,
0.079
1.113
0.024
0.377
0.926
0.011
0.148
0.049
0.988
R U decomposition.
V R decomposition.
Polar Decomposition
0 Comments
http://www.continuummechanics.org/polardec...
Continuum Mechanics
Recommend 3
Login
Sort by Oldest
Share
Thank You
Thank you for visiting this
webpage. Feel free to email
me if you have questions.
Also, please consider visiting
an advertiser on this page.
Doing so helps generate
revenue to support this
website.
9 of 10
Privacy
Polar Decomposition
http://www.continuummechanics.org/polardec...
Bob McGinty
bmcginty@gmail.com
Table of Contents
10 of 10